Introduction

Mortality among people with chronic obstructive pulmonary disease (COPD) is high at 30 000 deaths annually.1 Besides COPD itself, cardiovascular events are among the most common contributors to death among patients with COPD, often responsible for more deaths than respiratory events.2 Moreover, comorbid cardiovascular disease (CVD) is highly prevalent in patients with COPD, affecting between 5% and 46% of patients, depending on the CVD.3 4

The relationship between inhaled corticosteroids (ICSs) and cardiovascular events in people with COPD is debated. Some randomised controlled trial (RCT) post hoc analyses,5 observational studies,6 7 and meta-analyses8 suggest that people with COPD-prescribed ICSs may have reduced risk of cardiovascular events and cardiovascular-specific mortality. Equally, other meta-analyses have demonstrated no association between ICSs and cardiovascular events.9 10 Mixed results are likely due to (1) heterogeneity of study populations and study designs (such as ICS dosages, drug classes and duration that a patient is on therapy) and (2) the role of ICS being more complex than a simple blanket-effect. For example, ICSs have been shown to be protective against all-cause mortality, however, the effect was stronger after 6 months of treatment, at low or medium dosage and when the ICS class was budesonide.11 Additionally, patient cardiovascular history may modify the ICS-cardiovascular association where patients with COPD prescribed ICSs may have reduced cardiovascular event risk but CVD-naïve patients with COPD may not.8

We, therefore, aimed, in a representative COPD population to understand the relationship between ICS prescriptions and major adverse cardiovascular events (MACEs; acute coronary syndrome (ACS), heart failure (HF), ischaemic stroke (stroke) or cardiovascular-specific death (CV-death)). We explored treatment regimens (mono and combination therapies), ICS drug class-subtyping, new usership and patient cardiovascular history.

MethodsData source and sample

A COPD cohort was defined using primary care records from the Clinical Practice Research Datalink (CPRD) Aurum database (May 2022 build).12 CPRD Aurum data are deidentified electronic healthcare records routinely collected from general practitioner (GP) practices in the UK, using EMIS Web software. CPRD data cover approximately 20% of the UK population (mostly England) and are representative of age, sex and region.13 Aurum primary care data were linked with secondary care Hospital Episode Statistics (HES) data,14 Index of Multiple Deprivation (IMD) data15 to determine socioeconomic status and Office of National Statistics death certifications.16 Full data information and linkage details and practices are in the online supplemental file 1.

Inclusion criteria were (1) a diagnostic code for COPD (the code list for COPD diagnostic codes was made using previously validated methodology17 and is available on GitHub: https://github.com/NHLI-Respiratory-Epi/ICS-MACE-in-COPD), (2) 40 years old or above, (3) current or ex-smokers, (4) had data recorded in CPRD Aurum from the 1 January 2010 to 31 December 2019, (5) registered at GP practice for at least 1 year prior to start of follow-up, (6) record quality deemed by CPRD as acceptable for research and (7) had been prescribed either ICSs or a long-acting bronchodilators.

Study design

We conducted an observational intention-to-treat cohort study. The start of follow-up was the latest date of a person’s COPD diagnosis, 40th birthday, a year after their GP registration date and 1 January 2010. The end of follow-up (in the absence of outcome) was the earliest date of 31 December 2019, GP practice or CPRD out-transfer or death (figure 1, online supplemental fi gure E1). We repeated this study design for incident treatment only where the start of follow-up among new users was the date that they initiated their inhaler treatment, within 1 year of meeting the initial inclusion criteria. The end of follow-up among new users (in the absence of outcome) was 1 year after initiating treatment or the earliest date of 31 December 2019, GP practice or CPRD out-transfer or death (if before the 1 year mark) (figure 1).

Figure 1

Study design for whole cohort design and new user cohort design. COPD, chronic obstructive pulmonary disease; CPRD, Clinical Practice Research Datalink; GP, general practitioner; MACE, major adverse cardiovascular event.

Exposure

The exposure was ICS prescription for COPD, defined in primary care, in the year preceding follow-up. We had two sets of exposures: one general exposure (‘any ICS’ vs ‘any non-ICS’) and one specific exposure (triple therapy vs combination (dual) long-acting bronchodilators). For the general exposure, any ICS was defined as at least one prescription for any ICS monotherapy, ICS-LABA (long-acting beta-agonist), ICS-LAMA (long-acting muscarinic-antagonist) or triple therapy (ICS-LABA-LAMA). Any ‘non-ICS’ was defined as a prescription for any LABA-monotherapy, LAMA-monotherapy and/or dual therapy (LABA-LAMA) and the absence of any ICS prescription in the year preceding follow-up. For the specific exposure, triple and LABA-LAMA therapies were defined as at least one prescription in the year preceding follow-up. Following these comparisons, we split and compared ICS (and triple therapy) by ICS-subtype consisting of a ‘fluticasone-group’ (fluticasone propionate or fluticasone furoate) and an ‘MBBC-group’ (mometasone furoate, beclomethasone, budesonide or ciclesonide).

Outcome

The outcome was a MACE including ACS, ischaemic stroke, HF or CV-death, defined in either primary (CPRD Aurum) or secondary care (HES).

Statistical analysis

Baseline characteristics were recorded for the overall cohort (online supplemental table E1) and subgroups (online supplemental table E2, E3) as mean (SD) for continuous data and counts (%) for categorical data. Baseline characteristics (code lists: https://github.com/NHLI-Respiratory-Epi/ICS-MACE-in-COPD) were taken nearest to the start of the follow-up date and again to a new start of follow-up for the new user cohort. We included age, sex, smoking status (ex or current), body mass index (BMI, most recent within 5 years, underweight: <18.5 kg/m−2; healthy weight: 18.5–24.9 kg/m−2; overweight: 25.0–29.9 kg/m−2; obesity: >30.0 kg/m−2), socioeconomic status (IMD quintiles; a measure of deprivation at Lower Super Output Area level where IMD1 is least deprived and IMD5 is most deprived), acute exacerbations of COPD group (year preceding follow-up, defined according to a previously validated algorithm where moderate exacerbations were managed in primary care and severe exacerbations were managed in secondary care), COPD severity (spirometry-derived global initiative for chronic obstructive lung disease (GOLD) group, 2 years preceding or first 3 months following follow-up start), short-acting bronchodilators (year preceding follow-up) and any record of the following covariates occurring at any time prior to follow-up: Asthma, depression, anxiety, gastro-oesophageal reflux disease (GORD), lung cancer, hypertension, diabetes and CVD history (ACS, HF or stroke; recorded in primary care or secondary care).

We implemented Bonferroni-corrected multivariate Cox proportional hazard regressions, adjusted for all covariates (selected based on prior clinical knowledge), to investigate the association between ICSs and MACEs. To address confounding by indication, we repeated analyses using doubly robust propensity score-adjusted Cox proportional hazard regression. Furthermore, where the exposure violated the proportional hazards assumption, we ran fully-adjusted time-interaction Cox models and calculated hazard ratios (95% CI) for each year of follow-up. We also conducted sensitivity analyses, including: (1) HES-defined MACEs only and (2) patients with no evidence of asthma after exploratory analyses showed large differences in ICS prescriptions between groups. Analyses were performed using Stata V.17 software.

We compared (1) any ICS prescriptions with any non-ICS, (2) triple therapy with LABA-LAMA, (3) subtyped-ICS groups with non-ICS, (4) subtyped-triple therapy with LABA-LAMA, (5) new ICS users with non-ICS new users and (6) triple therapy new users with LABA-LAMA new users. For each comparison ((1)–(6)), we repeated the analysis on CVD history-naïve patients only (online supplemental figure E2).

Results

Our study consisted of 113 353 patients with COPD: 77 787 (68.6%) were prescribed some form of ICS while 35 566 (31.4%) were not prescribed ICSs but were prescribed a long-acting bronchodilator (online supplemental table E1, figure E1). Patients with COPD prescribed ICSs were similar in age (mean=67.9 years) and sex (53.3% male) to those not prescribed ICSs. Smoking history, BMI, depression, anxiety, GORD, lung cancer, diabetes, cardiovascular history, short-acting bronchodilator medication and exacerbation frequency and severity were similarly distributed among ICS and non-ICS groups. More patients prescribed ICSs had evidence of asthma (current and/or history) (46.9%) than non-ICS patients (12.8%) (online supplemental table E1). Kaplan-Meier plots are available in the supplement (online supplemental figure E3).

Propensity scores, time-interaction analyses and sensitivity analyses

Propensity score-adjusted Cox model results were no different from traditionally-adjusted Cox models. Therefore, we report results from traditionally adjusted Cox models in this manuscript. Sensitivity analyses using HES-derived outcomes yielded similar results with one exception (discussed in ‘ICS and MACE subtypes’ below). Excluding patients with evidence of asthma yielded similar results as when patients with evidence of asthma were included. Crude models, propensity score-adjusted models and sensitivity analyses are available in the supplement (online supplemental table E4-E9).

HF and CV-death outcomes tended to violate the proportional hazards assumption, particularly for triple therapy versus LABA-LAMA where event numbers were lower (see supplement for annual or monthly time-interaction estimates online supplemental table E10,E11).

Fluticasone-ICS proportions consisted of fluticasone propionate (97.3%) and fluticasone furoate (2.7%) while MBBC-ICS proportions consisted of mometasone furoate (0.1%), beclomethasone (72.1%), budesonide (27.4%) or ciclesonide (0.4%) (online supplemental table E12).

ICSs and MACE

Within all patients with COPD (n=113 353), compared by ICS or non-ICS prescriptions, across a mean follow-up of 4.67 (SD=3.29) years, there were 18 999 MACEs and no significant association between ICSs and MACEs (adjusted HR (95% CI), HR (95% CI) = 0.98 (0.95, 1.02), p=0.41) (figure 2A). Among CVD-naïve patients (n=94 358), there were 11 659 MACEs and no significant association between ICSs and MACEs (HR (95% CI)=0.97 (0.93, 1.01), p=0.15) (figure 2B).

Figure 2

Comparison of MACE in patients with COPD on ICS therapy compared with non-ICS therapy (A+B) and on triple therapy compared with LABA-LAMA (C+D). Bold=statistically significant at p<0.01 (Bonferroni corrected). Any ICS=prescription in the year preceding follow-up of ICS monotherapy, ICS-LABA, ICS-LAMA or ICS-LABA-LAMA. Non-ICS=prescription in the year before follow-up of LABA monotherapy, LAMA monotherapy or LABA-LAMA. Triple=any combination prescription in the year preceding follow-up of ICS-LABA-LAMA. LABA-LAMA=any combination prescription in the year preceding follow-up of LABA-LAMA. ACS, acute coronary syndrome; COPD, chronic obstructive pulmonary disease; CVD, cardiovascular disease; CV death, cardiovascular-specific death; HF, heart failure; ICS, inhaled corticosteroid; LABA, long-acting beta-agonist; LAMA, long-acting muscarinic-antagonist; MACE, major adverse cardiovascular event.

Within ICS-subtypes, there was no significant association between MACEs and fluticasone-ICSs (HR (95% CI)=0.99 (0.96, 1.03), p=0.98) or MBBC-ICSs (HR (95% CI)=0.97 (0.94, 1.01), p=0.18) compared with non-ICSs (figure 3A). In CVD-naïve patients, there was no significant association between MACEs and fluticasone-ICSs (HR (95% CI)=0.98 (0.94, 1.03), p=0.50) nor MBBC-ICSs (HR (95% CI)=0.96 (0.94, 1.03), p=0.08) (figure 3B).

Figure 3

Comparison of MACE in patients with COPD on any ICS fluticasone or any ICS MBBC compared with non-ICS therapy (A+B); and on fluticasone triple therapy or MBBC triple therapy compared with LABA-LAMA (C+D). Bold=statistically significant at p<0.005 (Bonferroni corrected). ICS fluticasone=ICS is either fluticasone furoate or fluticasone propionate. ICS MBBC=ICS is either mometasone furoate, beclomethasone, budesonide or ciclesonide. Any non-ICS=prescription in the year before follow-up of LABA monotherapy, LAMA monotherapy or LABA-LAMA. Triple fluticasone=ICS component is either fluticasone furoate or fluticasone propionate. Triple MBBC=ICS component is mometasone furoate, beclomethasone, budesonide or ciclesonide. LABA-LAMA=any combination prescription in the year preceding follow-up of LABA-LAMA. ACS, acute coronary syndrome; COPD, chronic obstructive pulmonary disease; CVD, cardiovascular disease; CV death, cardiovascular-specific death; HF, heart failure; ICS, inhaled corticosteroid; LABA, long-acting beta-agonist; LAMA, long-acting muscarinic-antagonist; MACE, major adverse cardiovascular event; MBBC, mometasone furoate, beclomethasone, budesonide or ciclesonide.

Finally, we found no significant association between ICS new users and MACEs (HR (95% CI)=1.13 (1.04, 1.24), p=0.01 (Bonferroni corrected)) compared with non-ICS new users (figure 4A). Among new user CVD-naïve patients, we found no significant association between ICSs and MACEs (HR (95% CI)=1.12 (0.98, 1.28), p=0.11) (figure 4B).

Figure 4

Comparison of MACE in patients with COPD who are new ICS users compared with new long-acting bronchodilator users (A+B); and new triple therapy users compared with new LABA-LAMA users (C+D). Bold=statistically significant at p<0.01 (Bonferroni corrected). Any ICS=prescription in the year preceding follow-up of ICS monotherapy, ICS-LABA, ICS-LAMA or ICS-LABA-LAMA in the absence of a prior prescription. Non-ICS=prescription in the year before follow-up of LABA monotherapy, LAMA monotherapy or LABA-LAMA in the absence of a prior prescription. Triple=any combination prescription in the year preceding follow-up of ICS-LABA-LAMA in the absence of a prior prescription. LABA-LAMA=any combination prescription in the year preceding follow-up of LABA-LAMA in the absence of a prior prescription. ACS, acute coronary syndrome; COPD, chronic obstructive pulmonary disease; CVD, cardiovascular disease; CV death, cardiovascular-specific death; HF, heart failure; ICS, inhaled corticosteroid; LABA, long-acting beta-agonist; LAMA, long-acting muscarinic-antagonist; MACE, major adverse cardiovascular event.

ICS and MACE subtypes

There was a statistically significant HF reduction among people prescribed ICSs compared with non-ICSs (HR (95% CI)=0.91 (0.86, 0.96), p<0.001) (figure 2A), although the proportional hazards assumption was violated. When modelled with time interaction, the annual HF association was significantly reduced until year 6 of follow-up and was not significantly different than non-ICSs thereafter (figure 5) (online supplemental table E10). ICSs were also associated with a statistically significant reduction of HF in CVD-naïve people (HR (95% CI)=0.89 (0.83, 0.95), p<0.001) (figure 2B). Regarding ICS-subtypes, there was a significant reduction of HF in all patients prescribed MBBC-ICSs (HR (95% CI)=0.89 (0.84, 0.94), p<0.001) and in CVD-naïve patients prescribed MBBC-ICSs (HR (95% CI)=0.87 (0.81, 0.94), p<0.001) but not for fluticasone-ICSs (figure 3A,B). However, when defined in HES only, there was no significant association between ICSs and HF (online supplemental table E4) but the significant result for MBBC-ICSs remained (online supplemental table E5). Finally, there was no association between patients newly prescribed ICSs and HF compared with patients newly prescribed long-acting bronchodilators (non-ICSs) whether CVD-naïve or not (figure 4A,B).

Figure 5

Annual comparisons of heart failure in patients with COPD on any ICS therapy compared with long-acting bronchodilators, over the follow-up period, incorporating time interaction. Gold line=average HR across follow-up (as per original Cox model). Any ICS=prescription in the year preceding follow-up of ICS monotherapy, ICS-LABA, ICS-LAMA or ICS-LABA-LAMA. Non-ICS=prescription in the year before follow-up of LABA monotherapy, LAMA monotherapy or LABA-LAMA. Triple therapy=any combination prescription in the year preceding follow-up of ICS-LABA-LAMA. LABA-LAMA therapy=any combination prescription in the year preceding follow-up of LABA-LAMA. Specific numbers (as well as other time-varying models that were not statistically significant on average) available in online supplemental table E10. COPD, chronic obstructive pulmonary disease; ICS, inhaled corticosteroid; LABA, long-acting beta-agonist; LAMA, long-acting muscarinic-antagonist.

We found no association between ICSs and ACS in all patients, CVD-naïve patients or among any ICS-subtype group. However, people prescribed ICSs had a significantly increased rate of ACS as new users (HR (95% CI)=1.27 (1.09, 1.47), p=0.002) but not in CVD-naïve new users, specifically (figure 4A,B).

Compared with non-ICS users, whether CVD-naïve or not, we found no significant association between ICSs and stroke or CV death; in the whole cohort (figure 2A,B), by ICS-subtype (figure 3A,B) or in new users (figure 4A,B).

Triple therapy and MACE

There was no significant association between triple therapy and MACEs compared with LABA-LAMA, whether CVD-naïve or not, across the whole cohort (figure 2C,D), by ICS-subtype (figure 3C,D), new usership (figure 4C,D) or CVD-naivety.

Triple therapy and MACE subtypes

There was no significant association between triple therapy and any individual MACE compared with LABA-LAMA across the whole cohort (figure 2C,D), by ICS-subtype (figure 3C,D), new usership (figure 4C,D) or CVD-naivety.

Discussion

There is largely no association between ICSs and MACEs in people with COPD whether compared generally as ICSs versus long-acting bronchodilators (non-ICSs) nor when specifically comparing triple therapy to LABA-LAMA. ICS subtypes, new usership and patient cardiovascular history did not alter findings. ICS therapy may be HF-protective specifically compared with long-acting bronchodilators and the relationship appeared driven by MBBC-ICSs. The ICS-HF protective relationship existed until year 6 of follow-up and was, thereafter, no different to long-acting bronchodilators. However, given this association was only seen when including both primary and secondary care-defined HF events, the association is likely driven by a reduction in misclassification of COPD exacerbation events. ICS prescription was associated with increased ACS compared with those newly on long-acting bronchodilators but not beyond new usership.

Contextualisation with previous research

There is ample but heterogeneous literature among RCTs and observational studies on inhalers and cardiovascular events in patients with COPD: Studies have investigated different combinations, dosages, stages of inhaler use and cardiovascular outcomes.8 Not all studies address confounding by indication where patients who experience frequent or severe exacerbations—and are, therefore, possibly sicker—may be prescribed ICS, creating unbalanced study groups. Finally, studies have not investigated temporal differences in the association of ICS therapy and cardiovascular events.

Previous post hoc RCT analyses,5 observational research6 7 and meta-analyses8 indicated that ICSs may have a cardioprotective effect while other meta-analyses9 18 have not, the latter aligning with our findings. Drug class subtyping did not alter findings. We showed that HF risk was reduced by the budesonide-containing group which was also previously demonstrated with ischaemic heart disease in a post hoc analysis of the European Respiratory Society's study on chronic obstructive pulmonary disease (EUROSCOP) 3-year placebo-controlled study.19 An Italian cohort20 and a Swedish population-based registry study (the PATHOS study)21 found patients on budesonide/formoterol had fewer COPD exacerbations requiring healthcare than patients on fluticasone/salmeterol,21 possibly due to reduced pneumonia risk.22 Fewer COPD exacerbations could reduce consequential cardiovascular risk from resultant inflammation.

We found no difference in MACE rates between triple therapy and LABA-LAMA, supported by previous observational research, including within new users.23 Conversely, in a subanalysis of the phase III randomised, double-blind, multicentre efficacy and safety of triple therapy in obstructive lung disease (ETHOS) trial, triple therapy was associated with a reduced rate of CV-death.24 25 However, the number of CV deaths in the trial was low and trial populations often produce atypical disease populations due to balancing or participant selection, particularly when the original trial aim is not specific to the subanalysis.

Potential explanations of findings

The ICS-MACE hypothesis is that ICSs reduce COPD-related systemic inflammation and mitigate inflammation-resultant cardiovascular events. However, we demonstrated no association between ICSs and MACEs, possibly due to ICSs not suppressing systemic inflammation enough, not inhibiting specific mechanistic pathways or only suppressing inflammation within the lungs and not systemically.26 There is also an increased risk of pneumonia among people prescribed ICSs;27 hence, any benefit from ICS-related reduced systemic inflammation could be neutralised by immunosuppression, resulting in increased susceptibility to pneumonia-related COPD exacerbations, in turn, increasing MACEs.

MACE-subtyping allowed examination of the types of cardiovascular events that ICSs may be associated with and we found an association between ICSs and HF (first 5 years of follow-up) but not with ACS, stroke or CV-death. Patient phenotype or cardiovascular event type may, therefore, influence whether ICSs are cardioprotective. The COPD-HF relationship is well-established where comorbid COPD and HF reciprocally worsen prognosis.28–31 Effectively treating one disease may reduce the rate at which disease reciprocity occurs. For example, previous work from our group demonstrated that prevalent ICS use reduced lung function decline rates in COPD.32 Treatment of COPD lung function decline and exacerbations by ICSs may reduce HF progression due to treating COPD itself, rather than ICS addressing HF directly. The relationship between HF and ICSs was also associated with ICS subtype where HF was significantly reduced by MBBC-ICSs but not by fluticasone-ICSs. Previous research demonstrated reduced pneumonia risk in patients with COPD prescribed budesonide or beclomethasone compared with those prescribed fluticasone.22 27 It may be that the magnitude of the positive association between pneumonia (and potentially, therefore, pneumonia-related COPD exacerbations) and fluticasone ICSs neutralises any cardioprotective benefit whereas among patients prescribed budesonide or beclomethasone-ICSs where there is a comparatively weaker pneumonia association, the downstream association with HF remains.

Secondary care-defined HF was not associated with ICSs indicating potential misclassification in primary care due to the overlap of symptoms of COPD and HF. As ICSs treat COPD exacerbations, people on ICSs could have fewer coded HF events due to fewer COPD exacerbations that may have been incorrectly coded as HF (as a result of symptomatic overlap). Additionally, high doses of corticosteroids, typically prescribed orally to patients with COPD during exacerbations, increase HF.33 The discrepancy could be due to the systemic effects of oral corticosteroids at very high doses and over a short period rather than inhaled and taken at lower doses for longer. Cumulative dose ICSs, therefore, may not be cardioprotective which may explain why ICSs were only associated with reduced HF up to year 6 of follow-up.

Previous research34 and this research showed no association between ICSs and myocardial infarction (the majority of our ACS definition) but we showed one minor exception where new users of ICSs had increased ACS rates compared with new users of long-acting bronchodilators. Although reasons for this are somewhat unclear, one supposition is that new ICS prescriptions may have been in response to a step-up in COPD which increased ACS risk (ie, the new ICS prescription is an artefact of another unmeasured COPD-related or CVD-related factor that immediately preceded prescription). Despite statistically correcting for multiple tests, it is also possible this finding has arisen due to chance and requires further investigation.

Limitations and strengths

Perhaps our biggest limitation is the potential misclassification of disease as our outcome included both primary care and hospitalisation events, resulting in possible event rate overestimation or underestimation.35 Although validation research has demonstrated that MACEs (eg, HF) can be identified using our methods (positive predictive value (PPV)=80%),36 our HES-only sensitivity analysis was not statistically significant and, hence, the ICS-HF relationship remains unclear. Although we time-updated covariates in new users (follow-up was 1 year later), we did not time-update covariates across follow-up. Although most covariates were chronic and unlikely to change, we acknowledge that some may change (such as exacerbation group). As this study is an intention-to-treat design, it is possible that some individuals may have changed medication type from their pre-follow-up grouping at some point during follow-up. Furthermore, as this is an observational study using electronic healthcare records, medication allocation is not random; clinicians prescribe therapy according to patients that they perceive will benefit from it. We chose only to investigate two binary comparisons (ICS vs non-ICS and triple therapy vs LABA-LAMA), therefore, we do not know whether specific ICS/LABA/LAMA combinations are associated with MACEs. We chose to retain statistical power to investigate multiple aspects of MACEs and ICSs generally with the intention of highlighting areas for which more granular studies could be done. Finally, MACE definitions vary across different studies; deciding which elements to include within MACE definitions is largely dependent on the study question. ICSs have already been shown to reduce all-cause mortality11 and as our study outcome was cardiovascular-specific, we elected to include cardiovascular-specific mortality to avoid introducing non-CVD-related noise into the data. We, however, acknowledge that there is a competing risk for all-cause mortality.

Our study has several strengths. We investigated numerous aspects of the ICS-MACE relationship (ICS-subtypes, new usership, cardiovascular history and MACE-subtypes) within one COPD population from a database that is approximately representative of the UK population and with high completeness of diagnostic coding. Our methods have been used and validated in other research23 including how we identified our cohort,17 designed code lists37 and generated algorithms38 for covariates. We undertook sensitivity analyses to cross-check findings for HES-defined MACEs only and to assess outcomes in asthma-free patients only. In addition to covariate adjustment, we conducted a sensitivity analysis adjusting for propensity scores to address confounding by indication. Our results were almost identical indicating the robustness of our findings. Finally, we tested the proportional hazards assumption in all models and—where violated—we included time as an interaction, calculated the HRs across each time increment and showed how the exposure effect changed across follow-up.

Future directions and clinical implications

We demonstrated that there is largely no cardioprotective effect of ICSs. While we investigated multiple aspects of ICS therapy, two aspects that require additional research include ICS dosage and time on ICSs. Second, an important question arising from this research is whether medication response is influenced by demographics or COPD phenotypes. Previously, the protective effect of ICS against coronary heart disease was demonstrated to be particularly strong in older adults, younger men and former smokers.39 ICS was also associated with reduced all-cause mortality in patients with COPD with high eosinophil counts, a history of more than two exacerbations within the previous year and a more severe GOLD stage.11 Our research demonstrated some evidence of cardioprotection against HF but not other MACEs. Understanding whether COPD patient phenotypes have fewer MACEs due to responding well to a particular medication protocol is essential for managing patient outcomes.

Data availability statement

Data may be obtained from a third party and are not publicly available. Data sets generated and/or analysed in this study are not publicly available, however, data are available on request from the CPRD. Their provision requires the purchase of a license and this license does not permit the authors to make them publicly available to all. This work used data from the version collected in May 2022 and has clearly specified the data selected in the Methods section. To allow identical data to be obtained by others, via the purchase of a license, the code lists will be provided upon request. Licenses are available from the CPRD (http://www.cprd.com): The Clinical Practice Research Datalink Group, The Medicines and Healthcare products Regulatory Agency, 10 South Colonnade, Canary Wharf, London E14 4PU.